State Preparation and Initialisation
State preparation resets qubits to a known state and builds the desired input; it is a DiVincenzo criterion and a real error source.
Starting from a known state
Every quantum computation begins with initialisation: resetting the qubit register to a known, fixed state, conventionally |000...0>. Reliable initialisation is one of the DiVincenzo criteria, because an algorithm's correctness assumes a clean starting point.
How reset is done
- Passive: let qubits relax to their ground state |0> over several T1 times
- Active: measure each qubit and apply an X if it reads 1 (measurement-based reset)
- Cooling: for some platforms, sideband or algorithmic cooling drives qubits to the ground state
Passive relaxation is simple but slow; active reset is faster but inherits any measurement error. The quality of initialisation is captured in state-preparation-and-measurement (SPAM) error, a headline hardware metric.
Building the input state
From the clean |000...0>, gates construct whatever input the algorithm needs. Hadamards create uniform superpositions; controlled gates and rotations build structured or entangled inputs. Preparing an arbitrary n-qubit state is itself hard — general amplitude encoding can require a number of gates exponential in n — so efficient state preparation is an active area of algorithm design.
Why it limits everything
Errors in preparation propagate through the entire circuit. If the initial state is already impure, no downstream gate can fully recover the intended computation. This is why SPAM error is reported alongside gate fidelity, and why benchmarking methods like randomized benchmarking are designed to separate gate error from preparation error.
In error correction
Fault-tolerant machines constantly prepare fresh ancilla and logical states — for instance, high-fidelity |0> and special magic states needed for non-Clifford gates. Preparing these reliably, through repeated measurement and distillation, is a substantial part of the overhead of quantum error correction.